Traditional systems may compute similarity between two media assets based on metadata attributes. For example, the system may use a model by which individual media assets are considered similar based on shared metadata attributes. Although the similarity metrics produced by these systems may be effective, the models do not take into account other factors that can improve the similarity metrics.